Draft version April 25, 2014

A Preprint typeset using L TEX style emulateapj v. 5/2/11

COMPLEXITY ON DWARF GALAXIES SCALE: A BIMODAL DISTRIBUTION FUNCTION IN SCULPTOR

Maarten A. Breddels and Amina Helmi
Kapteyn Astronomical Institute, University of Groningen and P.O. Box 800, 9700 AV Groningen, The Netherlands Draft version April 25, 2014

arXiv:1404.5958v1 [astro-ph.GA] 23 Apr 2014

ABSTRACT In previous work we have presented Schwarzschild models of the Sculptor dSph, demonstrating that this system could be embedded in dark matter halos that are either cusped or cored. Here we show that the non-parametric distribution function recovered through Schwarschilds method is bimodal in energy and angular momentum space for all best tting mass models explored. We demonstrate that this bimodality is directly related to the two components known to be present in Sculptor through stellar populations analysis, although our method is purely dynamical in nature and does not use this prior information. It therefore constitutes independent conrmation of the existence of two physically distinct dynamical components in Sculptor and suggests a rather complex assembly history for this dwarf galaxy. Subject headings: galaxies: dwarf galaxies: kinematics and dynamics
1. INTRODUCTION

Sculptor was the rst dwarf spheroidal (dSph) galaxy discovered (Shapley 1938), and it may be considered an archetype of the class of dSph as it is a rather featureless system (compared to e.g. Fornax, de Boer et al. 2011). Nonetheless, Scl has proven to be more complex than originally thought. Photometric surveys starting from Da Costa (1984), (and later by Light 1988; Hurley-Keller et al. 1999; Majewski et al. 1999), have revealed that its horizontal branch morphology changes with radius. More recently, thanks to large spectroscopic surveys, it has been possible to relate the dierences in the spatial distribution of the blue and red horizontal branch (BHB/RHB) to dierences in chemical composition and kinematics of its stars (Tolstoy et al. 2004), as well as to age gradients (de Boer et al. 2011). The picture that has emerged form this body of work is that Sculptor has two populations or components. The rst population is centrally concentrated, metal rich, younger, is represented in RHB stars and has colder kinematics, with a decreasing line-of-sight velocity dispersion with distance from the centre. The second population is less concentrated, metal poor, older, is prominent in BHB stars, has hotter kinematics and a more constant velocity dispersion prole. It was not clear until now whether this could simply be due to a population gradient. dSph have been the target of many kinematic surveys in the past decade because of their very high dynamical mass-to-light ratios. The aim of these studies has been to provide constraints on the nature and distribution of dark matter, however no rm conclusions have yet been drawn. This is because of limitations in the data (only access to line-of-sight velocities) and also in the models. For example, the most-widely used modeling technique is based on the Jeans Equations and requires making assumptions about the orbital structure of the system (see e.g. Walker et al. 2009b), although recent work by Richardson & Fairbairn (2013, 2014) usbreddels@astro.rug.nl

ing higher moments and the virial equations, seems to be able to circumvent this degeneracy. A more powerful approach is to use orbit-based dynamical modeling, also known as Schwarzschilds method. Breddels et al. (2013) have shown that the data on Scl does not constrain the inner slope of its dark matter density prole very strongly, and that neither cored nor cuspy (of the NFW-type, Navarro et al. 1996) proles are favoured. On the other hand, it has been argued that the multiple components present in Sculptor should be used to model this system dynamically and that this ought to lead to much tighter constraints. This is because these components are hosted by the same underlying potential and their presence eectively would reduce the available parameter space of plausible dynamical models (Battaglia et al. 2008). Several studies have attempted this using e.g. the virial equations (Agnello & Evans 2012) or constraints based on Jeans modeling (Walker & Pe narrubia 2011; Amorisco & Evans 2012b), and concluded that NFWlike proles are strongly disfavoured. All these works have assumed that the two populations are splitted in the same way in photometry and kinematics/chemistry, although this is a priori not warranted, as the kinematics (and split in metallicity) are obtained from the RGB while the photometry is t for the BHB and RHB stars independently. Perhaps more importantly, models based on a single estimate of the mass at a given radius (e.g. either through Jeans or through the virial equations) have assumed the components to follow the same functional form of the light distribution (with dierent characteristic parameters). As we shall show below, this is not necessarily a valid asssumption. In this Letter we analyse the phase-space structure of the best tting Schwarzschild dynamical models of Sculptor from Breddels & Helmi (2013) and Breddels (2013). Our spherical orbit based dynamical models are nonparametric in the distribution function (e.g. no assumptions are made on the anisotropy) and provide good ts to the global kinematics and light distribution of Scl. The dark matter distribution follows specic parametric

2
discrete, nfw moments, nfw

moments, core13

moments, core24

Fig. 1. Velocity dispersion (top panel) and kurtosis (lower panel) for the best tting models for dierent dark matter proles, namely Navarro, Frenk & White (NFW; based on moments or discrete modeling of the data) and two cored proles, with 2 )2 (core24). (r ) 1/(r + rc )3 (core13) and (r ) 1/(r 2 + rc

Fig. 2. Orbit-weights for the best t models for dierent dark matter halo proles: discrete NFW (Breddels 2013, top left); moment-based NFW (top right) and cored (bottom) models from Breddels & Helmi (2013). The red dashed and blue solid contours indicate the two components found by the watershed method.

proles (c.f. Jardel et al. 2012), which all produce very similar mass distributions in a nite region around the half-light radius of Scl. As we shall show below, the orbit weights (which correspond to the distribution function of the galaxy) show a bimodal distribution for all the best t models of Sculptor, even though this is not a priori assumed.
2. DISTRIBUTION FUNCTIONS OBTAINED VIA
SCHWARZSCHILD MODELS

Our orbit based dynamical models are described extensively in Breddels et al. (2013). This method essentially consists in nding a linear combination of orbits (integrated in a given gravitational potential) that allows tting the kinematic data and light distribution of the system. By varying the characteristic parameters of the gravitational potential (e.g. mass and scale radius), best ts are found, while the weights of the orbits eectively provide the distribution function non-parametrically. Fig. 1 shows how well the best t models for various dark matter proles explored in Breddels & Helmi (2013) and from the discrete modeling approach by Breddels (2013, PhD thesis) reproduce the velocity dispersion and kurtosis proles. As can be seen from this gure the resulting ts are nearly indistinguishable from one another. Fig. 2 shows the orbit-weights that dene the distribution functions of the best t models explored. For example, the best t NFW model found in Breddels (2013, PhD thesis), obtained using discrete tting is given in the upper left panel. The associated distribution function appears to be bimodal as it has two distinct components: one at low angular momentum (near energy index 8, angular momentum index 2), and a second one

at higher angular momentum (near energy index 7, angular momentum index 6). This bimodality is also present in the distribution functions of the remaining best tting models from Breddels & Helmi (2013), whether cored or cusped. Although the exact location of the lower angular momentum component varies slightly, it is reassuring that the bimodality is found using both the discrete and the traditional Schwarzschild methods and also for dierent dark matter proles. It demonstrates the robustness of the result as well as the power of using non-parametric Schwarzschild methods. To obtain further insights into the nature of these two components we use a watershed algorithm. We rst locate the local maxima in the image, which serve to identify the two dierent components. Starting from the pixel with the maxima, we follow the landscape downwards, i.e. we nd those neighbouring pixels that have a lower value. These are associated to either of the components, until no more such pixels are found. At this point we exclude pixels that could be simultaneously associated to both components. In Fig. 2 we indicate the pixels associated to each of the two components by showing the contour that correspond to 5% of the maximum orbital-weight for each component. We then construct two distribution functions by selecting the pixels identied by the algorithm as just described. We now refer to the higher and lower angular momentum components as the blue and red components respectively.
3. THE NATURE OF THE BIMODALITY

From the distribution function, we can derive the line of sight velocity as well as the surface brightness proles for each component. This is useful in order to understand the link to the cold/hotter kinematic or RHB/BHB pop-

3
discrete, nfw moments, nfw discrete, nfw moments, nfw

moments, core13

moments, core24

moments, core13

moments, core24

Fig. 4. Density of the projected light distribution for the two components, similar to Fig. 3. Solid black graph indicates the total light distribution. Thicker and semi-transparant lines indicate the best tting Plummer prole.

Fig. 3. The black points with errorbars shown the velocity dispersion prole for Scl. The top four panels show the los for each of our best tting models, separately for the low angular momentum (dashed red) and high angular momentum components (dotted blue) as identied in Fig. 2. The bottom panel shows the velocity dispersion proles as function of elliptical radii for the metal rich and metal poor populations in Sculptor, using data from Battaglia et al. (2008) and also from Walker et al. (2009a) for the outer most bins (see the text for more details).

ulations known to exist in Sculptor. In Fig. 3 we show the line of sight velocity dispersion proles for the two components separately, where each panel corresponds to the best t models shown in Fig 2. The red component shows an increasing velocity dispersion towards the centre and falls o rapidly with radius, while the blue component has a atter prole. These trends are quite similar to the velocity dispersion proles shown at the bottom of the gure for the MR and MP components dened as those stars with [Fe/H] > 1.5 dex (red) and [Fe/H] < 1.7 dex (blue) in Battaglia et al. (2008). In this gure we have also added the data from Walker et al. (2009a) in the four outermost bins for a fairer comparison to the data shown in the top panels and used for the Schwarzschild models. Note that the sharp [Fe/H] cuts used in this panel to dene the two populations do not guarantee the absence of cross-

contamination, especially in the inner regions, where the populations overlap spatially. In fact, any misassignment in the center will tend to lower the dispersion of the MR population, while increasing that of the MP, as perhaps these comparisons suggest. In Fig. 4 we show the projected light distribution for each of the two identied components (in red and blue, for low and high angular momenta respectively), as well as for the full system (in black). This gure clearly shows that the red component is more centrally concentrated than the blue component for all models. This can be quantied further by tting a Plummer prole to each component separately. The results are given in Table 1. Note however, that the Plummer functional form does not t well the light distribution of the more concentrated, lower angular momentum component (which is somewhat better t with an exponential). Typically the ratio of Plummer scale radii is bred /bblue 0.5, which is comparable to the ratio determined by tting Plummer proles to the RHB and BHB populations separately, which is r = 0.6 according to Battaglia (2007). However, recall that the photometric and even the spectroscopic decomposition do not necessairly correspond to the dynamical components found by our method. Nonetheless, the similarities are striking. In the above analysis, parts of the distribution function were not assigned to either of the two components (e.g. typically in the region in between). None of the conclusions we have reached so far are very sensitive to whether this intermediate mass is assigned to the red or the blue components. The fractional masses for these can be found in the right columns of Table 1, where we refer to the fraction that does not belong to either component as intermediate.

4
nfw(discrete) nfw core13 core24 bred 0.19 0.18 0.18 0.18 bblue 0.34 0.35 0.38 0.37 btotal 0.3 0.3 0.3 0.3 fred 0.080 0.088 0.097 0.069 fint 0.109 0.020 0.045 0.129 fblue 0.812 0.892 0.858 0.802

TABLE 1 Plummer scale radii b for the red (low angular momentum) and blue (high angular momentum) components, and for the total light distribution. The last three columns give the mass fractions associated to the red, blue and the intermediate components.

4. DISCUSSION AND CONCLUSIONS

We have shown that the distribution function of Sculptor for several best tting mass models is bimodal in energy and angular momemtum space. The two components may be split in a low and high angular momentum parts using the watershed method. The properties of the low and high angular momentum components are similar to the metal rich and metal-poor components respectively, known to be present in this galaxy, in terms of their velocity dispersion prole and their light distribution (Battaglia et al. 2008). This result is quite remarkable since we have not assumed at any point the existence of multiple components in the Sculptor dwarf. It therefore suggests that the metal-rich and metal-poor stars indeed are dynamically distinct, and that Sculptor is not simply a system with a radial gradient in stellar populations. This nding highlights the full power of the Schwarzschilds dynamical approach, and would not have been possible if we had taken a parametric approach to model the distribution function. The fact that our models naturally recover the bimodality present in Sculptor, for all dark matter proles

explored, and even for the NFW form, would seem to be at odds with the results of Agnello & Evans (2012) and Walker & Pe narrubia (2011). In these works, Sculptor was modeled as a two-component system with light distributions that followed similar proles. The use of the virial equations or the robust estimator of the mass at the half-light radii of each component was used to argue that NFW proles could be ruled out with high signicance. Our modeling however, shows this is not the case and why. The assumption of similar light proles appears to be crucial to reach those conclusions, and would seem to introduce a systematic bias that none of these works have taken into account. In the resulting non-parametric dynamical models we have obtained, the light distribution for the two components is quite dierent and cannot be parametrized well by Plummer proles as Fig. 4 shows. It is natural to wonder whether other dwarfs also depict multiple dynamical components. Fornax would be a natural candidate but its complex light distribution (e.g. the presence of shells, non-axisymmetries in the centre, Battaglia et al. 2006) and the hints of misaligned kinematics (see e.g. Amorisco & Evans 2012a) require the use of a more general (non-spherical) Schwarzschild modeling approach. On the other hand, it would be desirable to have larger datasets for e.g. Carina and Sextans to be condent in the robustness of the analyses. More generally, an interesting challenge will be to understand how such complex systems can form on the smallest galaxy scale, i.e. on the scale of the dwarf spheroidals.
ACKNOWLEDGEMENTS

AH and MB are grateful to NOVA for nancial support. AH acknowledges nancial support the European Research Council under ERC-Starting Grant GALACTICA-240271.

REFERENCES Agnello, A., & Evans, N. W. 2012, ApJ, 754, L39 Amorisco, N. C., & Evans, N. W. 2012a, ApJ, 756, L2 . 2012b, MNRAS, 419, 184 Battaglia, G. 2007, PhD thesis, Kapteyn Astronomical Institute, University of Groningen Battaglia, G., Helmi, A., Tolstoy, E., Irwin, M., Hill, V., & Jablonka, P. 2008, ApJ, 681, L13 Battaglia, G., et al. 2006, A&A, 459, 423 Breddels, M. 2013, PhD thesis, Kapteyn Astronomical Institute, University of Groningen Breddels, M. A., & Helmi, A. 2013, ArXiv e-prints Breddels, M. A., Helmi, A., van den Bosch, R. C. E., van de Ven, G., & Battaglia, G. 2013, MNRAS, 433, 3173 Da Costa, G. S. 1984, ApJ, 285, 483 de Boer, T. J. L., et al. 2011, A&A, 528, A119 Hurley-Keller, D., Mateo, M., & Grebel, E. K. 1999, ApJ, 523, L25 Jardel, J. R., Gebhardt, K., Fabricius, M., Drory, N., & Williams, M. J. 2012, ArXiv e-prints Light, R. M. 1988, PhD thesis, Yale University., New Haven, CT. Majewski, S. R., Siegel, M. H., Patterson, R. J., & Rood, R. T. 1999, ApJ, 520, L33 Navarro, J. F., Frenk, C. S., & White, S. D. M. 1996, ApJ, 462, 563 Richardson, T., & Fairbairn, M. 2013, MNRAS, 432, 3361 . 2014, ArXiv e-prints Shapley, H. 1938, Nature, 142, 715 Tolstoy, E., et al. 2004, ApJ, 617, L119 Walker, M. G., Mateo, M., & Olszewski, E. W. 2009a, AJ, 137, 3100 Walker, M. G., Mateo, M., Olszewski, E. W., Pe narrubia, J., Wyn Evans, N., & Gilmore, G. 2009b, ApJ, 704, 1274 Walker, M. G., & Pe narrubia, J. 2011, ApJ, 742, 20